TL;DR
- Testing Results: The Guardian found OpenAI’s GPT-5.2 cited Grokipedia nine times across tests on Iran and Holocaust-related topics.
- Source Quality Concerns: Grokipedia has been criticized for citing neo-Nazi forum Stormfront 42 times and white nationalist site VDARE 107 times.
- Industry-Wide Issue: Multiple AI models including Anthropic’s Claude have also been reported citing Grokipedia, indicating a broader problem with LLM source validation.
The Guardian has found OpenAI’s GPT-5.2 is citing Grokipedia, Elon Musk’s AI-powered encyclopedia known for referencing neo-Nazi sources, when users ask about Iran and the Holocaust.
OpenAI introduced GPT-5.2 in December 2025, positioning it as “the most capable model series yet for professional knowledge work.” In testing, GPT-5.2 cited Grokipedia nine times across more than a dozen questions on sensitive topics.
Tests Reveal Critical Citation Pattern
The model cited Grokipedia on queries about political structures in Iran, including salaries of the Basij paramilitary force and ownership of the Mostazafan Foundation. GPT-5.2 also referenced Grokipedia when asked about Sir Richard Evans, a British historian who served as expert witness against Holocaust denier David Irving.
When citing Grokipedia, GPT-5.2 repeated stronger claims about Iranian government links to telecom provider MTN-Irancell than those appearing on Wikipedia. This information filtered into responses on more obscure topics where the model apparently lacked competing sources.
Promo
The pattern reveals GPT-5.2’s source filters prioritize availability over reliability when facing knowledge gaps. This creates vulnerability to misinformation on niche topics lacking authoritative coverage.
Despite these citations, the model did not cite Grokipedia when prompted directly to repeat misinformation about the January 6 insurrection, media bias against Donald Trump, or the HIV/AIDS epidemic. The inconsistent behavior suggests GPT-5.2 may be applying content filters to certain politically sensitive topics while allowing Grokipedia citations through on others.
Grokipedia’s Problematic History
To understand why these citations raise concerns, Grokipedia is an AI-generated encyclopedia that launched in October 2025, aiming to compete with Wikipedia without allowing direct human editing. The platform has been criticized for propagating rightwing narratives on topics including gay marriage and the January 6 insurrection.
Research analysis found Grokipedia cites neo-Nazi forum Stormfront 42 times and white nationalist website VDARE 107 times. The platform also cites far-right hub Infowars 34 times.
Articles are often lifted from Wikipedia but introduce problematic sourcing and errors when they differ. The citation counts far exceed typical reference patterns for mainstream sources, indicating systemic reliance on fringe content.
“The guardrails are off. The publicly determined, community-oriented rules that try to maintain Wikipedia as a comprehensive, reliable, human-generated source are not in application on Grokipedia.”
Harold Triedman, computer science graduate and former senior privacy engineer for Wikimedia Foundation (via Futurism)
Industry-Wide Problem
This issue extends beyond OpenAI. GPT-5.2 is not the only LLM citing Grokipedia.
Anthropic’s Claude has also referenced it on topics from petroleum production to Scottish ales, according to anecdotal reports. Security experts raised concerns last spring that malign actors, including Russian propaganda networks, were seeding AI models with disinformation through a process called “LLM grooming,” deliberately poisoning the information ecosystem that language models draw from.
In June 2025, concerns were raised in US Congress that Google’s Gemini repeated the Chinese government’s position on human rights abuses in Xinjiang and China’s Covid-19 policies. The cross-platform citations indicate a structural flaw in how LLMs validate web sources during training and inference, positioning bad actors to exploit the entire AI ecosystem through strategic content placement.
Experts Warn of Credibility Contagion
Building on these systemic concerns, Nina Jankowicz, a disinformation researcher, described Grokipedia as “relying on sources that are untrustworthy at best, poorly sourced and deliberate disinformation at worst.” She warned of a credibility laundering effect where AI model citations legitimize questionable sources.
“They might say, ‘oh, ChatGPT is citing it, these models are citing it, it must be a decent source, surely they’ve vetted it’ – and they might go there and look for news about Ukraine”
Nina Jankowicz, disinformation researcher (via The Guardian)
AI models continued to cite a fabricated quote attributed to Jankowicz even after a major news outlet removed it. She noted that “most people won’t do the work necessary to figure out where the truth actually lies.”
The fabricated quote persistence demonstrates how LLM training creates permanent knowledge contamination that survives source corrections and requires costly retraining to eliminate.
OpenAI Defends Sourcing Approach
In response to these concerns, an OpenAI spokesperson stated the model “aims to draw from a broad range of publicly available sources and viewpoints.”
The spokesperson added: “We apply safety filters to reduce the risk of surfacing links associated with high-severity harms, and ChatGPT clearly shows which sources informed a response through citations.”
OpenAI claims GPT-5.2 Thinking hallucinates 30% less than its predecessor on de-identified ChatGPT queries. In contrast, when contacted for comment, an xAI spokesperson responded: “Legacy media lies.”
